Presentation is loading. Please wait.

Presentation is loading. Please wait.

1 The role of structural prediction in rapid syntactic analysis Ellen Lau, Clare Sroud, Silke Plesch, Colin Phillips, 2006 PSYC 525 10-6-2010 Soondo Baek.

Similar presentations


Presentation on theme: "1 The role of structural prediction in rapid syntactic analysis Ellen Lau, Clare Sroud, Silke Plesch, Colin Phillips, 2006 PSYC 525 10-6-2010 Soondo Baek."— Presentation transcript:

1 1 The role of structural prediction in rapid syntactic analysis Ellen Lau, Clare Sroud, Silke Plesch, Colin Phillips, 2006 PSYC 525 10-6-2010 Soondo Baek

2 2 Pre-introduction She did not kiss Dana’s of the bride. Very fast detection of category violation (Neville et al, 1991)

3 3 Pre-introduction she did not kiss Dana’s of the bride. Very fast detection of category violation (Neville et al, 1991) Modulation of this effect by the strength of prediction for a certain syntactic category Although Erica kissed Mary’s mother,

4 4 Introduction Syntactic processing is very fast, beginning within 300ms after word onset But the nature of early syntactic processes largely remains unspecified

5 5 Introduction Early Left Anterior Negativity (ELAN) –Increased left anterior negativity in 100-250ms induced by grammatical category violations –Observed in very specific contexts unlike other ERP components e.g., The scientist criticized.. Max’s of proof the theorem. Max’s proof of the theorem.

6 6 Introduction Other ERP components associated with a variety of syntactic processes –Left Anterior Negativity (LAN) in 300-500ms inflection/agreement violation, some case violation, word category violation, violation of constratins on wh- fronting, syntactic garden-path –P600 (Syntactic Positive Shift) in 500-900ms category violation, agreement violation, syntactic garden-path, subcategorization violation ambiguity, long-distance dependecy construction, unexpected thematic role assignment

7 7 Introduction How can category violation be detected so fast? –Earlier account (e.g., Friederici, 1995) Only category information is avaliable at the earliest stage of structure building (e.g., Frazier, 1987) Syntactic structure building is generally fast and ELAN results when the category of a word cannot integrate to the preceding structure –New proposal tested Preceding contexts strictly constrain the possible category of an upcoming word, dramatically reducing possibilities to be considered Max’s [(ADJ) NOUN]

8 8 Introduction Ellipsis construction used to test the proposal –*He met John, but I met Mary’s. –The possessor Mary’s requires a noun phrase, leading to a strong prediction for an upcoming noun phrase –He met Jonh’s mother, but I met Mary’s. –The possessor Mary’s can be followed by a null element referring to mother, not necessarily predicting an upcoming noun phrase

9 9 Materials (32 items per condition) and predictions +Ellipsis Grammatical Ungrammatical -Ellipsis Grammatical Ungrammatical Agreement Grammatical Ungrammatical Although Erica kissed Mary’s mother, she did not kiss the daughter of the bride. she did not kiss Dana’s of the bride. Although the bridesmaid kissed Mary, she did not kiss the daughter of the bride. she did not kiss Dana’s of the bride. Although Matt followed the directions closely, Although Matt follow the directions closely, he had trouble finding the theater.

10 10 Materials Sentence completion test to verify that ellipsis manipulation works –Although Erica kissed Mary’s mother, she did not kiss Dana’s … –Although Erica kissed Mary, she did not kiss Dana’s …

11 11 Methods 32 participants Each word presented for 300ms 200ms blank between words Press buttons to judge acceptability

12 12 +AlthoughEricakissedMary’smother,shedidnotkissDana’softhebride.? Demonstration: +Ellipsis ungrammatical condition

13 13 +AlthoughthebridemaidkissedMary,shedidnotkissDana’softhebride.? Demonstration: -Ellipsis ungrammatical condition

14 14 Results: Pre-critical word Effect of grammaticality at pre-criticial words at the latency of 300-500ms –she did not kiss the daughter of the bride. –she did not kiss Dana’s of the bride.  This made it undesirable to directly compare grammatical and ungrammatical condition A significant main effect of ellipsis at pre-criticial word, but with small amplitude difference

15 15 Results: Critical word (of) Ungrammatical conditions (~ Dana’s of) 0-200ms inteval –Overall, slightly greater negativity in the +ellipsis condition than in the –ellipsis condition, but no difference at any individual region 200-400ms interval (ELAN) –Greater negativity at the left anterior region in the –ellipsis condition than in the +ellipsis condition –Slightly greater positivity at the right posterior region in the –ellipsis condition than in the +ellipsis condition 600-1000ms interval (P600) –No differences between +ellipsis and –ellipsis condition

16 16

17 17

18 18 Results: Critical word (of) Grammatical conditions (~ daughter of) 0-200ms interval –Slightly greater negativity at the right anterior region in the –ellipsis condition than in the +ellipsis condition 200-400ms interval –No effect on ellipsis at any individual region 300-500ms interval –Greater negativity at the right posterior region in the +ellipsis than in the –ellipsis condition Scarce reliable contrasts between thet two ellipisis conditions

19 19 Results: Agreement violation 0-200, 200-400, 300-500ms intervals –No effect of grammaticality 600-1000ms interval (P600) –Posterior positivity & anterior negativity

20 20 Discussion Earlier account of ELAN effect to category violations –reflects an initial stage of structure building –predicts that any kind of category violations should yield a response with a similar latency Alternative account –Eearly response to category violations reflects strong local prediction for an upcoming word

21 21 Discussion Results showed that –Left anterior negativity was reduced when preceding context allowed ellipsis interpretation of possesors (e.g., Dana’s) as compared with when there was no such context Not all category violations are diagnosed in the same fashion Strong local prediction may explain early detection of category violations, without need to appeal to an initial parse stage sensitive only to category information

22 22 Discussion Strong local prediction for an upcoming word may work in two ways to enable early detection of (category) violations –Detection of mismatch with prediction is simpler and faster than evaluation of all possible ways to attach an incoming word to the existing structure  But this predicts ELAN to grammatical sequences like ‘Max’s clever proof’, which was shown not to be the case in Austin & Phillips (2004) –Very restricted possibilities induced by preceding context enables faster detection of violations as compared with cases where more possibilities must be searched before decision

23 23 Discussion: things to note Comparisons within each level of grammaticality conditions –Traditional ELAN based on the difference between ungrammatical and grammatical condition –do not show whether the availability of ellipsis interpretation removed or just reduced ELAN –do not allow the test of P600 effect Delayed onset of ELAN at 200ms as compared with some studies that showed earlier onet at 100-150ms –(together with the presence of posterior positivity) makes it unclear whether the finding is really about ELAN


Download ppt "1 The role of structural prediction in rapid syntactic analysis Ellen Lau, Clare Sroud, Silke Plesch, Colin Phillips, 2006 PSYC 525 10-6-2010 Soondo Baek."

Similar presentations


Ads by Google